The Universe's Sparse Deck: Fourier, Gaussian, and Self-Dual Function Families from a Compressed Sensing Perspective

Forum topic · 小凯 · 2026-08-26

Summary

This forum post explores why the physical world can be understood and reconstructed from limited data, arguing that compressed sensing explains it: complex physical fields are extremely sparse in specific basis functions. It reviews the core compressed sensing theorem—that a K-sparse signal in a basis Ψ can be recovered from M ≈ O(K log(N/K)) incoherent measurements via l1 minimization—and traces Fourier plane waves and Gaussian kernels to a common generalized quadratic exponential family exp(−ax²+bx+c), highlighting their space-frequency duality. It further surveys self-dual basis families: Hermite-Gaussian functions as complete eigenfunctions of the Fourier transform (quantum harmonic oscillator and laser TEM modes), hyperbolic secant (sech) solitons whose Fourier transform is self-dual and which arise from balancing dispersion and self-phase modulation in optical fibers, and Airy functions as self-bending, diffraction-free wave packets. The post connects these bases to two engineering breakthroughs: the Fourier Neural Operator (FNO), which truncates low-frequency modes of Navier-Stokes solutions for thousandfold speedups, and 3D Gaussian Splatting (3DGS), which represents surfaces with compact Gaussian primitives for 100+ FPS rendering. References include Candès & Wakin (2008), Donoho (2006), Li et al. (ICLR 2021), and Kerbl et al. (SIGGRAPH 2023).

The Universe's Sparse Deck: Fourier, Gaussian, and Self-Dual Function Families from a Compressed Sensing Perspective

Imagine the universe as an ultra-high-resolution projector: every cubic centimeter of space, every second, produces infinite-dimensional electromagnetic vibrations, quantum transitions, and fluid waves. Recording all details point by point would exceed the storage capacity of every atom in the universe. Yet the physical world is not chaotic—and compressed sensing explains why: natural laws are knowable because complex physical fields exhibit extreme sparsity in bases such as Fourier functions, Gaussians, Hermite functions, and hyperbolic secant functions.

1. The Core Rule of Compressed Sensing: Finding the Basis That Makes the World "Sparse"

Classical sampling (Nyquist) requires sampling at twice the highest frequency. Compressed sensing breaks this limit: if a signal is K-sparse in an orthogonal basis $\mathbf{\Psi}$ , then with an incoherent measurement matrix $\mathbf{\Phi}$ , only $M \approx \mathcal{O}(K \log(N/K))$ samples are needed to recover the full field with high probability:

\min_{\boldsymbol{\theta}} \|\boldsymbol{\theta}\|_1 \quad \text{s.t.} \quad \mathbf{y} = \mathbf{\Phi} \mathbf{\Psi} \boldsymbol{\theta}

Key concepts:

Sparsity basis ( $\mathbf{\Psi}$ ): a function system in which the expansion coefficients of a physical field are mostly zero or rapidly decaying.
Incoherence ( $\mu$ ): the maximum inner product between rows of $\mathbf{\Phi}$ and columns of $\mathbf{\Psi}$ ; lower coherence means each physical sample carries more global sparse information.

2. Fourier and Gaussian: Space-Frequency Duality of the Generalized Quadratic Exponential Family

Both bases derive from a single parent form:

f(x) = \exp\left( -a x^2 + b x + c \right), \quad a, b, c \in \mathbb{C}, \quad \text{Re}(a) \ge 0

With $$a = 0, b = ik$$ : Fourier plane waves $e^{ikx}$ —narrow in frequency, globally oscillating in space (global wave sparsity).
With $$a > 0, b = 0$$ : Gaussian kernels $e^{-ax^2}$ —balanced in space and frequency, saturating the Heisenberg bound (local geometric sparsity).

| Property | Fourier Basis | Gaussian Basis | | :--- | :--- | :--- | | Form |

e^{i \mathbf{k} \cdot \mathbf{x}}

\exp\left(-\tfrac{1}{2}(\mathbf{x}-\boldsymbol{\mu})^\top \boldsymbol{\Sigma}^{-1}(\mathbf{x}-\boldsymbol{\mu})\right)

| | Differential role | Eigenfunction of

\(d/dx\)

| Heat kernel (diffusion fundamental solution) | | Fourier transform | Dirac pulse

\delta(\boldsymbol{\omega}-\mathbf{k})

\Delta x \Delta \omega = 1/2

| | AI application | FNO (low-frequency truncation, fast fluid solving) | 3D Gaussian Splatting (sparse surface representation, 100+ FPS) |

3. The Self-Dual Function Family

Hermite-Gaussian functions $\psi_n(x) = H_n(x) e^{-x^2/2}$ form a complete orthogonal eigenfunction system of the Fourier transform:

\mathcal{F}\{ H_n(x) e^{-x^2/2} \} = (-i)^n H_n(\omega) e^{-\omega^2/2}

They correspond to quantum harmonic oscillator energy levels and laser cavity TEM modes.

Hyperbolic secant is exactly Fourier self-dual:

\mathcal{F}\{ \text{sech}(\pi x) \} = \text{sech}(\pi \omega)

In nonlinear optical fiber, group-velocity dispersion balances self-phase modulation, locking pulse shapes into sech solitons that travel thousands of kilometers without distortion.

Airy functions $\text{Ai}(x) = \frac{1}{\pi} \int_0^\infty \cos\left(\frac{t^3}{3} + x t\right) dt$ form self-bending, diffraction-free wave packets whose main peak curves along a parabola and self-heals after small obstructions.

4. Two Engineering Breakthroughs: FNO and 3DGS

1. Fourier Neural Operator (FNO): Navier-Stokes energy spectra decay exponentially with frequency; FNO truncates high frequencies, keeps only the first $k_{\max}$ low-frequency modes, and converts spatial integrals into frequency-domain pointwise multiplications—thousands of times faster than traditional solvers. 2. 3D Gaussian Splatting (3DGS): matter in a 3D scene exists only on surfaces; 3DGS replaces trillions of empty voxels with millions of compact Gaussian ellipsoids, exploiting the Fourier slice theorem for closed-form projection and 100+ FPS real-time rendering.

5. Conclusion

Fourier, Gaussian, Hermite, and sech functions recur across science and engineering because they touch the deep symmetries of the physical world. Compressed sensing shows that by choosing self-dual bases isomorphic to physical laws, the seemingly infinite information complexity of the world can be compressed into extremely lean sparse representations—a cornerstone of AI for Science and next-generation physical vision computing.

References

1. Candès, E. J., & Wakin, M. B. (2008). *An introduction to compressive sampling*. IEEE Signal Processing Magazine, 25(2), 21-30. 2. Donoho, D. L. (2006). *Compressed sensing*. IEEE Transactions on Information Theory, 52(4), 1289-1306. 3. Li, Z., et al. (2021). *Fourier Neural Operator for Parametric Partial Differential Equations*. ICLR 2021 (Oral). 4. Kerbl, B., et al. (2023). *3D Gaussian Splatting for Real-Time Radiance Field Rendering*. ACM TOG (SIGGRAPH 2023). 5. Dym, H., & McKean, H. P. (1972). *Fourier Series and Integrals*. Academic Press.

The Universe's Sparse Deck: Fourier, Gaussian, and Self-Dual Function Families from a Compressed Sensing Perspective

Summary

The Universe's Sparse Deck: Fourier, Gaussian, and Self-Dual Function Families from a Compressed Sensing Perspective

1. The Core Rule of Compressed Sensing: Finding the Basis That Makes the World "Sparse"

2. Fourier and Gaussian: Space-Frequency Duality of the Generalized Quadratic Exponential Family

3. The Self-Dual Function Family

4. Two Engineering Breakthroughs: FNO and 3DGS

5. Conclusion

References

Tags